101 results about "Bayesian neural networks" patented technology

The invention provides a method for predicting the residual life of a power transformer based on digital-analog fusion, and belongs to the technical field of transformer detection.The method comprises the steps that multi-dimensional sensor data of the power transformer is collected, wavelet transform preprocessing is conducted, a normalized data matrix is established, a physical equation is established, and a deterministic physical model is formed; a data-driven model is established based on an improved adaptive multi-scale network to realize multi-scale feature adaptive extraction, a topological phase change algorithm is introduced to identify key transition points in an aging process, and a deterministic physical model and the data-driven model are fused to establish a digital-analog fusion prediction framework. A generative adversarial network is adopted to perform data enhancement to solve the problem of scarcity of fault samples, a Bayesian neural network and a Monte Carlo random inactivation technology are utilized to construct an uncertainty quantization framework to output a residual life prediction value and a confidence interval thereof, and the technical problem that the prediction precision of the residual life of the transformer is not high is solved.

The invention provides a mechanical arm control method and system based on mechanism-Bayesian joint modeling, and relates to the technical field of robot control, and the method comprises the steps that firstly, a mechanical arm mechanism model is constructed, parameters of the mechanical arm mechanism model are estimated, and preliminary dynamics prediction is obtained; then, combining with motor driving torque observation data, establishing a random mathematical model of mechanism model residual errors, and decomposing the random mathematical model into deterministic and random parts; carrying out probability learning on the residual error by utilizing a Bayesian neural network, and outputting a prediction mean value and a variance of the residual error; in combination with preliminary dynamic prediction and residual information, constructing a data-driven uncertainty adaptive control law without dependence of an acceleration signal, and carrying out random stability analysis; and a stable joint driving torque instruction is generated, and high-precision trajectory tracking of the mechanical arm is achieved. According to the method, the interpretability of the mechanism model and the high adaptability of the data driving model are combined, the control precision and flexibility are effectively considered, and the robustness and reliability of the mechanical arm in the complex dynamic environment are improved.

The invention discloses a mechanical equipment residual service life prediction and uncertainty quantitative analysis method and device. The method comprises the following steps: acquiring multi-dimensional time sequence sensor data generated by mechanical equipment to be predicted in an operation process; inputting the multi-dimensional time sequence sensor data into a pre-trained physical constraint Bayesian neural network model; wherein the physical constraint Bayesian neural network model comprises a segmented bidirectional long short-term memory network feature extraction module, a hierarchical gating recursive regression network degradation dynamic modeling module and a Bayesian reasoning and physical constraint fusion regression module which are connected in sequence; the segmented bidirectional long-short-term memory network feature extraction module is used for processing input multi-dimensional time sequence sensor data and outputting hidden feature vectors representing local degradation dynamics of equipment; the hierarchical gating recursive regression network degradation dynamic modeling module is used for processing hidden feature vectors and capturing global degradation dynamic features through a complex numerical value hidden state updating mechanism; the Bayesian reasoning and physical constraint fusion regression module is used for carrying out Weibull distribution parameter regression based on the global degradation dynamic characteristics, introducing a deep implicit physical residual constraint and outputting a probability distribution parameter of the residual service life; and based on the probability distribution parameters, generating a residual service life prediction result and uncertainty quantitative information of the mechanical equipment.

The invention discloses a power load prediction method, and the method comprises the steps: firstly solving an extreme event data sparsity problem through a generative adversarial network, and constructing an event time sequence library through a time sequence anomaly detection algorithm; then analyzing the causal relationship between the event and the load by applying a causal discovery algorithm, and converting prediction output into probability distribution by adopting a Bayesian neural network to quantify uncertainty; constructing a prediction model triggered by an event, and generating a multi-time scale probability prediction interval; and finally, generating a multi-scene prediction result through Monte Carlo simulation, quantifying the system recovery capability in combination with a toughness index, and integrating the system recovery capability to a decision support system to generate a risk response scheme. According to the method, the accuracy and robustness of load prediction under the extreme climate are remarkably improved, full-chain risk insight from early warning to recovery is realized, and prospective decision support is provided for safe operation of a power system.

The invention provides an electric quantity prediction method and system fusing physical constraint factors, and relates to the technical field of electric quantity prediction. Historical load, weather, electricity price and calendar data are collected, and a key feature set is constructed through preprocessing and feature selection; a prediction model with the physical information neural network as the core is constructed, the prediction model comprises a recursion sub-module used for short-term prediction and a trend sub-module used for long-term prediction, and a physical constraint loss item based on a physical rule is introduced into model training so as to enhance the generalization ability; a multi-time granularity modeling framework is adopted, uncertainty quantization is achieved through a Monte Carlo Dropout or Bayesian neural network, and a confidence interval of a predicted value is output; and finally, causal reasoning is carried out through a Shapley value algorithm and anti-fact simulation, and key influence factors are identified. According to the method, the precision, stability and interpretability of electric quantity prediction are effectively improved, and reliable support is provided for power grid dispatching and decision making.

The invention discloses a super-set deterministic weather forecast method and device based on machine learning, and the method comprises the steps: obtaining multi-source meteorological data of a target region, carrying out the meshing of the multi-source meteorological data, carrying out the historical static feature analysis and dynamic feature analysis of the meshing features, obtaining the feature weight of each mode, and carrying out the recognition of the multi-source meteorological data. The method comprises the following steps: constructing a prediction sub-model according to an extreme event, obtaining enhanced numerical prediction data, obtaining posterior probability distribution of grid points through a conditional generative adversarial network and a Bayesian neural network based on the numerical prediction data, a gridding feature and a feature weight, taking a maximum probability value as a deterministic weather forecast, and calculating a confidence interval. And obtaining a joint probability product including wind speed and rainfall joint distribution and the characteristic contribution degree. According to the method, numerical forecasting set products of different mode centers are utilized to fuse probability forecasting information, deterministic weather forecasting is obtained, smoothing of extreme events is reduced, deterministic maximum value output is provided, and meanwhile good interpretability is achieved.

The invention provides a bridge life prediction method and system based on a physical information neural network and fusing a physical degradation mechanism and monitoring data, realizes reliable and real-time prediction of the residual life of a bridge, and relates to the technical field of bridge structure health monitoring. The method comprises the following steps: constructing a bridge real-time feature tensor, forming a bridge real-time feature tensor with uniform space-time alignment, constructing a physical information neural network model by taking a space-time coordinate (x, t) in the formed bridge real-time feature tensor as network input, and outputting endogenous physical field data for representing a degeneration state of a detected bridge; retraining the physical information neural network model; and taking endogenous physical field data output by the retrained physical information neural network model as input, and feeding the endogenous physical field data into a pre-trained Bayesian neural network model to realize uncertainty quantification and prediction of the service life of the tested bridge.

The invention provides a river flow online measurement and calculation method based on multi-source data fusion, and belongs to the technical field of river flow measurement. A Bayesian neural network and active learning collaborative hydrological memory reconstruction model is adopted to carry out high-confidence data interpolation on a sensor failure period and quantify uncertainty, fractional calculus is introduced to model a river memory effect, and flow evolution is analyzed through a fractional order water balance equation. And selecting a steady-state or non-constant flow calculation mode according to the flow change rate, inverting an optimal flow field for the non-constant flow by adopting a four-dimensional variational data assimilation method in combination with regularization constraint and time smoothing constraint, and outputting a flow measurement result and an uncertainty quantitative index. The technical problems that flow measurement and calculation data are missing and accurate interpolation is difficult due to sensor failure under the extreme hydrological condition are solved.

The invention relates to the technical field of artificial intelligence, and discloses a Bayesian neural network method fused with uncertainty quantization and a system thereof, the method comprises five steps of probability weight reconstruction modulo, variational posterior inference, re-parameterization sampling, multi-scale uncertainty quantization and adaptive rejection decision, the network weight reconstruction modulo is probability distribution, and the probability distribution of the network weight reconstruction modulo is improved. Sparse induction is realized by adopting scale Gaussian mixture prior, calculation complexity is reduced through a local re-parameterization technique, reasoning time consumption is controlled within two times of a deterministic network, a multi-scale uncertainty quantification module quantifies cognitive uncertainty by counting output variance of multiple forward propagation, and the accuracy of reasoning is improved. According to the method, the technical problem that a deep learning model lacks reliable prediction confidence estimation is effectively solved, and the misdiagnosis sample omission ratio in a medical image classification task is reduced.

The invention relates to the technical field of first-aid repair tower safety assessment, and discloses a first-aid repair tower safety risk assessment method and device, and the method comprises the steps: building a first-aid repair tower assessment model based on a Bayesian neural network, and enabling the first-aid repair tower assessment model to comprise an input layer node, a middle layer node and an output layer node, the input layer node represents a risk cause of the first-aid repair tower, the middle layer node represents a middle assessment node of the risk of the first-aid repair tower, and the output layer node represents a risk assessment target of the first-aid repair tower; acquiring first-aid repair tower data and environment data; inputting the first-aid repair tower data and the environment data into the first-aid repair tower evaluation model for probabilistic reasoning to obtain state probability distribution of the output layer nodes; and performing risk assessment on the first-aid repair tower according to the state probability distribution of the output layer nodes.

The invention relates to the technical field of wind turbine generator fault early warning, and discloses a wind turbine generator fault early warning method and system based on multi-modal data fusion, and the method comprises the steps: collecting the data of a multi-modal sensor, and carrying out the time-space alignment preprocessing; multi-modal features are extracted through variational mode decomposition, STL decomposition and other methods, and cross-modal fusion is achieved through dimension adaptive projection and a multi-head attention mechanism; calculating a dynamic weight based on three factors of data quality, fault type correlation and information gain, and carrying out weighted fusion; constructing a dynamic unit topological graph, and capturing cross-unit association features by using a space-time diagram convolutional network; long-time early warning with confidence is realized through double-branch gating fusion in combination with a Bayesian neural network; a multi-label classification identification multi-fault mode is adopted, and an operation and maintenance decision is optimized through an adaptive large neighborhood search algorithm. According to the method, the long early warning window of the offshore wind turbine generator can be realized, and uncertainty quantification and intelligent operation and maintenance decision support are provided.

The invention provides a false comment prediction method and system based on a Bayesian multi-scale attention network, and relates to the technical field of network risk prediction, and the method comprises the steps: obtaining network comment core data; feature extraction and covariable design are carried out on the network comment core data, and covariables and time sequence comment data are fused to form a feature matrix; inputting the feature matrix into a comment prediction model, learning an association relationship from different perspectives by using a plurality of attention heads, and generating a context enhancement feature vector containing cross-product information; inputting the context enhancement feature vector into a Bayesian neural network, and outputting a predicted mean value and a logarithm standard deviation of the number of false comments in each time unit; and converting the logarithmic standard deviation into a non-negative standard deviation through an activation function, constructing Gaussian probability distribution of the number of false comments, and realizing prediction uncertainty quantization. The prediction precision and the risk reference value are remarkably improved.

This invention relates to the field of aero-engine maintenance engineering and quality assessment technology, and discloses an aero-engine overhaul assessment method based on a physical information Bayesian neural network. The method includes collecting multi-source heterogeneous data from maintenance and testing sites and converting it into key process indicator scores; extracting rotor dynamics, imbalance transmission, and empirical formula features to construct physical constraint regularization terms; establishing an initial Bayesian neural network containing deterministic and variational Bayesian layers; inputting the key process indicator scores into the network, and optimizing parameters by combining the physical constraint regularization terms with adaptive annealing and early stop mechanisms; performing multiple Monte Carlo samplings on the trained network to calculate the mean distribution of the output results and obtain distribution statistical characteristics; calculating the quality score, confidence interval, and attribution warning information based on the distribution statistical characteristics, and outputting an overhaul assessment report. This invention solves the problem that existing pure data models violate physical laws and cannot quantify confidence levels.

The application discloses a joint cartilage stress dynamic monitoring system based on flexible sensing and AI and belongs to the technical field of medical health monitoring.The application solves the problem that the prior art can only measure single-point pressure or strain and cannot obtain full-field three-dimensional stress distribution of a cartilage contact surface, and the system function stops at monitoring and does not form a monitoring-to-warning closed loop, quantifies uncertainty through a Bayesian neural network, provides a reliable basis for clinical decision-making, combines multidimensional biomechanical characteristics and an attention mechanism, accurately focuses on a stress key area, significantly improves the precision and individualization level of joint cartilage health evaluation, helps early detection and prevention of injury, classifies stress abnormalities, implements accurate early warning, and formulates an individualized treatment scheme according to uncertainty results and cartilage health indexes, and customizes a rehabilitation training plan according to user conditions, and improves the scientific nature of joint injury prevention and rehabilitation.

The present application belongs to the field of geological exploration and artificial intelligence technology, and discloses a mineral resource prediction method and system based on a Bayesian physical information neural network. The method comprises the following steps: obtaining and preprocessing multi-source geological data of a study area; constructing a Bayesian neural network based on variational inference; constructing a physical constraint loss with a geological source term; performing three-stage progressive training based on a multi-objective loss function; and performing Monte Carlo sampling prediction and uncertainty quantification. The present application has strong physical interpretability: by embedding a steady-state diffusion equation with a geological source term, the model prediction result conforms to the geological law of ore-forming element migration and enrichment, and the source term clearly corresponds to two geological actions of fracture channel and ore-forming parent rock. The present application realizes the organic combination of data driving and physical driving by using multi-source information such as geochemical data, fracture structure, rock mass distribution and known mine point labels.

A method for training an artificial neural network, particularly a Bayesian neural network, using a training data set includes a step of matching the parameters of the artificial neural network according to a loss function. This loss function includes a first term representing an estimate of the lower bound of the distance between the classification of the training data set by the artificial neural network and the desired classification of the training data set. Furthermore, the loss function includes a second term configured to adjust for differences in random uncertainty in the training data set by different samples passed through the artificial neural network.

The application discloses a kind of based on active learning and man-machine collaborative optimization's massaging robot control method, it is related to artificial intelligence, medical rehabilitation robot and complex contact type operation control cross technical field, this method relies on multimodal perception system, core computing hub and remote man-machine collaborative interface control platform, in turn complete the initial massaging strategy model construction of multi-source heterogeneous expert data, the uncertainty measurement based on bayesian neural network, man-machine collaborative key frame active learning trigger, reinforcement learning reward function design of fusion patient biological feedback and incremental learning of impedance control parameter and force-position hybrid output;The application greatly improves the clinical safety of massaging robot, exponentially reduces data acquisition cost, realizes individual physiological closed-loop rigid-flexible massaging, also overcome the catastrophic forgetting problem of neural network, endow system lifelong learning ability.

This invention relates to the field of textile performance testing and intelligent evaluation technology, specifically providing a home textile sleep-aiding performance evaluation system and construction method based on feature function indicators. The system includes: acquiring five-dimensional feature indicators (tactile, thermal humidity, pressure, interference, and hygiene) through standardized instrument testing, and automatically assigning weights using range normalization and entropy weighting; constructing a physically constrained Bayesian neural network, embedding prior knowledge of materials science and sleep physiology as regularization terms into the loss function, and outputting a sleep-aiding performance score and confidence interval; establishing an adaptive weighted graph convolutional network to achieve knowledge transfer and zero-sample prediction between different home textile products; and fitting the relationship between static indicators and environmental parameters through a dynamic environment adaptive mapping module to output a scenario-based sleep-aiding performance level. This invention integrates instrumental quantitative testing with multi-level neural networks, breaking away from traditional linear regression dependence and achieving rapid, objective, personalized, and scenario-adaptive sleep-aiding performance evaluation.

The application discloses a kind of water turbine monitoring method and device, wherein the method comprises: using adaptive wavelet semi-soft threshold denoising method to the vibration signal data of water turbine is denoised;Characteristic data is extracted from vibration signal data using wavelet energy coefficient analysis method combined with wavelet decomposition coefficient mean square value statistical analysis method, and water turbine unit diagnosis sample database is established;Uncertainty bayesian neural network model based on deep learning is used to mine the feature relationship between water turbine data, which solves the problem that the existing technology detects water turbine by manual reinspection in offline state, with high implementation difficulty, low coverage, easy to miss detection, and unable to monitor water turbine unit in service state in real time.

The invention discloses a grounding equipment safety assessment method and system based on transfer learning, a medium and equipment. The method comprises the steps that environmental parameters of the environment where the grounding equipment to be evaluated is located are obtained, and the environmental parameters at least comprise the water content and the pH value of the soil environment where the grounding equipment is located; the environment parameters are input into a pre-trained corrosion prediction model, predicted value probability distribution of the corrosion rate of the grounding equipment is obtained, and the corrosion prediction model is obtained by conducting fine adjustment on a source domain pre-trained Bayesian neural network model through target domain data based on transfer learning; according to the predicted value probability distribution of the corrosion rate of the grounding equipment, calculating resistance change probability distribution of the grounding equipment through a preset physical model; based on the resistance change probability distribution of the grounding equipment, the probability that the resistance change value exceeds a preset safety threshold is calculated, and dynamic early warning is carried out based on the probability that the resistance change value exceeds the safety threshold, so that the prediction accuracy and reliability are improved.

The invention relates to a safety critical system risk dynamic assessment system based on multi-modal fusion and uncertainty quantification, and belongs to the technical field of industrial safety and artificial intelligence crossing. The technical problems of limitation of a single prediction model, lack of uncertainty representation, coarse and shallow multi-modal information fusion and the like of an existing safety evaluation system are solved. According to the method, a system architecture integrating multi-source heterogeneous data acquisition, space-time alignment preprocessing, modal adaptive fusion, uncertainty dynamic quantification and risk interpretability output is constructed. According to the system, an improved conformal prediction framework is adopted, and a Bayesian neural network and deep evidence learning are combined, so that real-time dynamic assessment and uncertainty accuracy measurement of safety critical system risks are realized. The core technology comprises multi-modal time sequence feature extraction of a gated cycle unit network based on an attention mechanism, fused hysteresis drift failure modeling of introducing a modal confidence imbalance coefficient and a modal missing detection delay coefficient, and a personalized uncertainty quantification method of conditional risk coverage guarantee. The method is especially suitable for the safety key fields such as industrial process control and infrastructure management, can provide double indexes of risk level and confidence for decision makers, and significantly improves the reliability and operability of risk assessment.

This invention provides a collision detection method and system for robotic arms based on Bayesian inference, relating to the technical fields of robot safety perception and human-computer interaction. The method includes: constructing a generalized momentum model containing unknown dynamic terms and external joint torques based on a joint spatial dynamics model of the robotic arm; performing probabilistic inference on the unknown dynamic terms using an integrated Bayesian neural network, outputting the predicted mean and variance of the unknown dynamic terms; constructing a data-driven adaptive momentum observer based on the predicted mean and variance to generate an estimated signal of the external joint torques; and performing collision detection based on sparse Bayesian inference on the estimated signal based on the sparsity characteristics of external collision perturbations to generate a collision determination signal. This method solves the technical problems of insufficient sensitivity and high false alarm rate in existing collision detection technologies, achieving improved accuracy in external torque estimation and collision detection sensitivity.

The invention discloses a smart tourism recommendation method based on granular computing and a type-2 fuzzy set. The method comprises the following steps: acquiring multi-source heterogeneous tourism data; performing uncertainty modeling on the multi-source heterogeneous tourism data by using a Bayesian neural network, outputting posterior distribution, and mapping the posterior distribution into membership function parameters of a generalized type-2 fuzzy set to obtain multi-modal type-2 fuzzy information particles; based on a particle calculation framework, with coverage rate-specificity collaborative maximization as a target, performing optimal particle size distribution on the multi-mode type-2 fuzzy information particles to generate optimal particle size information particles; inputting the optimal granularity information particles into a two-channel deductive learning framework, realizing knowledge-data dynamic fusion by the two channels through a shared attention layer, and outputting a fused tourist preference feature vector; and based on the preference feature vectors, constructing a multi-granularity graph neural collaborative filtering model, respectively generating a personalized recommendation list, a group recommendation list and a socialized recommendation list, and outputting interpretable rules.

The invention discloses an intelligent portfolio management method and system based on a Bayesian neural network, and belongs to the field of machine learning and financial investment management. In order to flexibly optimize the optimal investment portfolio under the conditions of risk avoidance and transaction cost change, the financial market investment income is maximized. By acquiring market financial data and processing information, a stochastic gradient variational Bayesian algorithm (SGVB) is adopted to solve the problem of uncertainty of stock income probability distribution parameters and an investment portfolio model structure, and parameter estimation errors are effectively solved. And then, the system combines the SGVB and a neural network to construct a probability deep learning framework, so that the model latent variables are modeled as probability distribution, and efficient investment portfolio model training is realized through stochastic gradient optimization. The method has outstanding feasibility and stability in the aspect of financial investment portfolio, reduces the transaction cost, improves the investment income, and adapts to a dynamic and complex financial market investment decision-making environment.

The invention relates to the field of ophthalmic disease image recognition, and discloses an ophthalmic disease image recognition method based on a neural network model, and the method comprises the steps: collecting a fundus photographic image and an OCT image, and carrying out the size standardization processing; outputting a high signal-to-noise ratio preprocessed image through a convolution attention module and a multi-scale noise suppression neural network; constructing a double-attention fusion framework; features are dynamically fused through a cross attention mechanism, and small sample rare disease recognition performance is optimized in combination with a meta-learning training strategy; a classification threshold value is dynamically adjusted through Bayesian neural network probability modeling, and a preliminary recognition result is output; semantic mapping is completed based on an ophthalmic clinical diagnosis term library, and a structured report containing focus coordinates, lesion grading and differential diagnosis bases is generated; the problems that in the prior art, the requirement for large-scale crowd screening is difficult to meet, feature extraction accuracy is not high, feature expression is not comprehensive, the recognition accuracy is low, and a structured report cannot be output are solved.

The invention provides a construction method of a multi-source heterogeneous equipment collaborative energy scheduling model, a medium and a system, and belongs to the technical field of energy scheduling. A fault propagation prediction model based on a heterogeneous graph structure and an attention mechanism is constructed, and the uncertainty is quantitatively predicted by using a Bayesian neural network; solving the collaborative scheduling mathematical model by adopting a self-adaptive robust optimization method to obtain an initial scheduling scheme, dynamically adjusting charging and discharging power according to a comprehensive aging index, guiding active learning through a sampling priority value to carry out dense sampling on high-uncertainty nodes, and carrying out incremental training on the model; and adjusting the power limitation of the energy storage unit and the control strategy of the photovoltaic inverter according to the node fault probability distribution, thereby solving the technical problem of insufficient reliability of the scheduling scheme caused by the lack of an artificial intelligence model training mechanism with high data efficiency in the fault propagation scene of the multi-source heterogeneous equipment of the micro-grid system.

The invention provides a physical AI virus identification method and system based on a Bayesian neural network, and belongs to the technical field of network security. According to the method, aiming at the fact that a target recognition object is a physical AI virus (including a physical confrontation sample, a physical backdoor, an unknown physical domain AI virus and the like), the physical confrontation sample is taken as an example, and the confrontation sample and a clean sample collected in a physical real world are creatively adopted to construct a training data set. According to the recognition framework provided by the invention, a multi-modal data fusion mechanism under simulation of real environment disturbance is utilized, higher robustness and generalization ability are shown in a complex physical scene, and AI viruses such as physical adversarial samples in the real world can be better recognized.

This invention proposes a method and system for detecting the operational status of a controlled fusion device, belonging to the field of controlled fusion device operational status detection technology. The method includes: obtaining a pulse height spectrum based on measurement results from a liquid scintillator detector; normalizing the count values ​​for each channel address; inputting the normalized pulse height spectrum data into a Bayesian neural network model for forward propagation calculation. The training process includes: constructing a neutron response matrix by combining Monte Carlo simulation and experimental calibration; generating a set of simulated neutron energy spectra based on preset basis functions; generating a set of simulated pulse height spectra by convolving the neutron response matrix with the set of simulated neutron energy spectra; training a neural network model with weight parameters following a probability distribution using Bayesian inference based on data pairs; and performing an inverse transform on the model output to obtain the incident neutron energy spectrum distribution. This invention improves the stability and accuracy of neutron energy spectrum interpretation and provides quantitative information on the uncertainty of the inversion results.

A training method, apparatus, electronic device, and storage medium for a Bayesian neural network based on a memristor array are disclosed. The conductance values ​​of the memristors in the memristor array are used to map the weights of the Bayesian neural network. The training method includes: acquiring first and second prior knowledge of the memristor array based on the inherent non-ideal characteristics of the memristors; calculating the total loss function of the Bayesian neural network based on the first prior knowledge; performing backpropagation on the total loss function to update the current parameters in the Bayesian neural network to obtain object parameters; and constraining the object parameters based on the second prior knowledge to obtain the training result of the weights of the Bayesian neural network. This training method can improve the robustness of the network to fluctuations in the memristor conductance values.

The application provides a section basin tracing method based on a deep Bayesian neural network, and comprises the following steps: acquiring real-time monitoring data of different sections; performing data analysis and data synthesis on the real-time monitoring data to obtain an initial data set and generate an input data set; designing a deep Bayesian neural network based on expert knowledge and experience; inputting the input data set into the Bayesian neural network and training and reasoning the Bayesian network; selecting a pollution source contribution rate as an evaluation index according to a tracing target, outputting the evaluation index as an evaluation result of the deep Bayesian neural network, obtaining a pollution source contribution degree of different monitoring factors in different pollution sources, and realizing the tracing of the pollution source.